Abstract

Every year, various natural disasters such as earthquakes and hurricanes strike our planet that account for over hundred-thousand casualties and hundred-millions affected across the globe. While destroying buildings and bridges, such natural phenomena cause roads to be blocked and areas to be unreachable, electricity to be unavailable, to name a few consequences among others. Besides that, the affected population needs to be transferred to medical centers and/or sheltered, provided with food, water, electricity, and other primary needs. Such tremendous demand for housing and emergency supplies is usually more than available resources. Hence, effective pre- and post-disaster logistics activities are essential in order to reduce the number of casualties. Motivated by the importance of preparedness planning in improving emergency logistics, this thesis is focused on the problem of locating temporary shelters and allocation of affected population to those shelters in the context of earthquakes. The aforementioned problem is formulated as two-stage stochastic and robust optimization models so as to incorporate the uncertain frequency, epicenter and magnitude of earthquakes that directly impacts the housing demand as well as availability of shelters and the transportation network. Along with introducing various corrective actions to hedge the preparedness plan against different earthquake scenarios, we also consider Conditional-value-at-Risk as the risk measure in the two-stage stochastic formulation. The idea is to protect the plan against a certain percentage of worst-case scenarios. More specifically, the robust model would make sure that the number of affected population that cannot be transferred to shelters (and consequently, the number of casualties) is minimized under a given confidence level. A case study inspired by a real earthquake is also designed that provides the opportunity to validate shelter location/allocation models proposed in this thesis. Finally, we run a set of computational experiments in order to compare the performance of deterministic, stochastic, and robust optimization models.